Researchers propose LLM-and-RL system for autonomous construction planning and progress updates
A new arXiv pre-print describes a method that combines a large language model with an enhanced reinforcement learning algorithm to plan crane actions, extract construction-environment information, and generate progress reports.
1 source · cross-referenced
- A pre-print proposes an autonomous construction system that pairs a large language model with reinforcement learning to plan crane actions and generate progress reports.
- The method uses an enhanced soft actor-critic reinforcement learning algorithm to plan mobile crane actions based on construction tasks.
- A customized embedding model extracts construction-environment information aligned with user interests.
- Feasibility was tested via full-scale retaining wall construction experiments.
Researchers from [Yifei Xiao, Huasheng Lin, T.Y. Yang] describe a system that integrates a large language model with reinforcement learning to automate construction tasks. The core method uses an enhanced soft actor-critic reinforcement learning algorithm to plan actions for a mobile crane based on specified construction tasks. Robot kinematics and control algorithms are then applied to translate those planned actions into executable crane movements. To align information extraction with user needs, the team developed a customized embedding model that processes construction-environment data according to user-specified interests.
The system also leverages a large language model—identified in the paper as ChatGPT 5—to update construction progress, deliver safety information to the control system, and automatically generate construction progress reports. The authors report that the feasibility of the proposed method was verified through several full-scale retaining wall construction experiments.
The work is presented as a response to growing housing demand and the limitations of traditional construction methods, positioning machine learning as a driver of smarter, more efficient building processes. The authors emphasize the role of large language models in enabling this transformation toward smart construction.
- Oct 21, 2026 · arXiv cs.AI
Multi-objective reinforcement learning method improves Pareto-optimal RTL circuit generation with LLMs
Trust79 - Oct 19, 2026 · arXiv cs.CL
Few prompt tokens drive most F1 loss when reducing experts in MoE prefill
Trust79 - Oct 19, 2026 · arXiv cs.CL
Study finds language models partially align with typological patterns but diverge in complement systems and action nominal constructions
Trust79